اصلاح و تایید در طراحی مدل جریان کار علمی پویا مبتنی بر حل مسئله

A science process is a process to solve complex scientific problems which usually have no mature solving methods. Science processes if modeled in workflow forms, i.e. scientific workflows, can be managed more effectively and performed more automatically. However, most current workflow models seldom take account of specific characteristics of science processes and are not very suitable for modeling scientific workflows. Therefore, a new workflow model named problem-based scientific workflow model (PBSWM) is proposed in this paper to accommodate those specific characteristics. Corresponding soundness verification and dynamic modification are discussed accordingly based on the new modelling method. This paper makes three main contributions: (1) three new constructs are proposed for special logic semantics in science processes; (2) verification is deployed with the consideration from both data-specific perspective and control-specific perspective; and (3) a set of rules are provided to automatically infer passive modifications caused by other modifications.

With the fast development of computer technologies, very large-scale complex science processes which cannot be deployed in the past can now be explored. A science process can be viewed as a process to solve a scientific problem with goals such as knowledge discovery, knowledge innovation, and so on. A large-scale complex science process needs to handle very complicated logics in scientific problems. Besides, in many cases, such a process involves a great number of scientists from different domains as well as distributed resources. To some extent, deploying such a science process is a huge engineering which requires automatic and reasonable management to the whole process. Fortunately, workflow technologies [24] provide useful ideas and paradigms for the automation of science processes. Workflow was originally emerged for the automation of a business process in whole or part, where a business process means a kind of process in the domain of business organizational structure and policy. The definition of a workflow consists of a coordinated set of activities that are connected in order to achieve a common goal. These activities can be organized in various routing forms such as sequential, parallel, and so on. Many workflow instances may be created and executed based on a workflow definition, and this process is deployed automatically under the control of workflow engines [14]. Many ideas and concepts from workflow technologies can be adapted to science processes, such as the idea of organizing a process as a workflow; the idea of using conditions to control automatic transitions between activities; the concept of building blocks, e.g. and-join, or-split, used for the representation of activity routings; and so on. More specifically, a science process can be modeled as a workflow before it is deployed. Such a workflow definition organizes the whole complex science process as a coordinated set of activities with data relationships and control relationships. The workflow should be verified according to certain rules or strategies. Then workflow engines control the deployment of the science process automatically based on the verified workflow. In this way, the science process can be deployed effectively and efficiently through a workflow form. Consequently, the scientific workflow is proposed to represent science processes in workflow forms, which can facilitate the automatic and effective performance of science processes. Though the scientific workflow roots in the traditional workflow i.e. workflow for automation of business processes, it is unsuitable to apply workflow models designed for business processes to the definitions of scientific workflows directly. This is because compared with business processes, science processes hold many characteristics which cannot be satisfied by current workflow models. Science processes differ a lot from business processes, where a science process is a tentative solving process without foregone stable solving schemes while a business process is a comparatively fixed procedure with a specific behavior description for each logic step. The detailed differences between business processes and science processes are analyzed as follows. First, science processes are more data-centric and knowledge-intensive [18], and require more powerful computing ability and mental thinking behaviors. In contrast, business processes pay more attention to control specifications than data specifications. Most activities seem like fixed mechanical operations. Second, activities in science processes perhaps are indeterminate at the definition stage because mature solving schemes usually do not exist. Therefore, different from business processes, science processes generally cannot be formalized as a stable series of ordered activities with specific behavior and fixed input or output specifications. Third, science processes are often highly creative, innovative and dynamic. The predefined scientific workflow models are often modified a lot at run-time execution stage, while business processes are comparatively stable. Fourth, a scientific workflow model is often modified at run-time execution stage because of indeterminacy and high dynamics. Hence, a scientific workflow model is high likely to be performed only one time while a business workflow model is usually executed repeatedly and has many instances. Fifth, participants of a science process, i.e. scientists, play an important role in the whole science process and maybe are the key factor to decide whether a scientific problem can be solved successfully. On the contrary, what participants do in business processes is comparatively simple interactive behaviors. Due to essential characteristics of science processes analyzed above, science processes are hardly organized in traditional workflow forms which consist of definite and concrete descriptions on each step, behavior, or task. Here, these traditional workflow forms mean Petri Nets based models [1], [9], [15] and [16], UML based models [5], [8] and [11], and so on. It is necessary to design a new workflow model which can really accommodate characteristics and laws of science processes. Unfortunately, although more and more researchers have realized the significance of applying workflow technologies to the deployment of large-scale complex science processes, most of their efforts have been put on the development of special scientific workflow management systems for science processes, such as Pegasus [10], Kepler [4], and GridNexus [7]. Up to now, design of scientific workflow models fully satisfying characteristics of science processes has not attracted enough attention. This paper is an endeavor to design a workflow model satisfying specific characteristics of science processes. Based on the analysis of the problem logic, a problem-based scientific workflow model (PBSWM) is proposed. Corresponding soundness of models is defined and theorems for verification are presented. Moreover, dynamic modifications on models at run-time execution stage are discussed and implemented through provided primitives and inferring rules. The main contributions of this paper are threefold. First, the problem-based modelling idea satisfies characteristics of science processes to a great extent. Based on that idea, three new relationships are suggested for the representation of special logic semantics in science processes. Second, verification is deployed from both data-specific perspective and control-specific perspective. This avoids conflicts between data relationships and control relationships which are hardly detected through general verification methods considering only one aspect. Last, a set of primitives are presented for implementation of active modifications and a set of inferring rules are provided for automatic derivation of passive modifications. All of these assist scientists to deploy many and frequent modifications in an effective way. The remainder of this paper is organized as follows. Section 2 introduces an example to demonstrate the necessity and issues of modelling science processes as scientific workflows. Section 3 puts forward problem-based scientific workflow model (PBSWM) based on the analysis of problem logics. Section 4 defines the soundness of PBSWM and provides corresponding theorems for verification. Section 5 proposes primitives and specific rules to dynamically modify the predefined scientific workflow model during the performance; also, verification on dynamically modified part is explored. Section 6 describes a case study about a scientific workflow model of the example given in Section 2. In Section 7, evaluation and relevant comparison are discussed. Section 8 concludes the paper and presents future work.

In this paper, a new perspective for modelling science processes in the workflow form, i.e. to model scientific workflows by taking subproblems as basic elements, is proposed. In PBSWM (Problem-Based Scientific Workflow Model), besides general data and control relationships, three new relationships, i.e. the context data relationship, the binding control relationship and the force_end relationship, are proposed to represent special logic semantics in scientific problems. These new relationships are significant and valuable not only for modelling scientific workflows but also for design of other flexible models through providing interesting ideas and perspectives. Based on control relationships and data relationships, a scientific workflow is organized as a structured model consisting of logically connected subproblems which are decomposed from a complex scientific problem. Corresponding soundness of PBSWM is defined and four theorems are provided for verification. Moreover, to adapt to the dynamic characteristics of science processes, both active and passive modifications can be deployed on the predefined workflow through proposed primitives and rules. In this way, a remodeled workflow can be generated, which reflects the innovative considerations of scientists during the performance. The future work includes the improvement on the structure of a PBSWM, such as organizing it in a hierarchical form so as to obtain more effective management for the large complex problem. We also plan to design specific verification algorithms under the direction of theorems in this paper. Though the complex structure with multiple relationships in a PBSWM will bring great challenges, the graph theory can be applied to address this issue. Moreover, implementation of the performance of a PBSWM is also an issue to be addressed. It mainly concerns design of workflow engines with special functions which can handle the new proposed relationships.